Time series


Fortran must die
a general thread on these. I cycle back to this every few years (since almost no one on the board is interested). :)

One of the strange thing about time series regression to me (ARDL or VAR models or regression with ARIMA error) is when exactly they should or should not be used. They involve issues where X occurs now, Y happens sometimes in the future. But in virtually all cases we can measure, the impact of X on Y is likely to be observed after X. So the real question is when you should use time series or not.

Note this is a different this then serial correlation. You can probably just use HAC standard errors to address this although there are certainly more complex ways to do so. Well actually do HAC deal with the problem....or do you need the more complex solutions such as regression with ARIMA error.